On Fri, Dec 13, 2013 at 1:55 AM, Francesco Fusco <ffu...@redhat.com> wrote: > On 12/12/2013 09:20 PM, Jesse Gross wrote: >> >> Out of curiosity, did you try using crc32q? OVS data structures are >> already aligned to 8 bytes. It would also be interesting to know if a >> parallelized implementation is worthwhile, although my guess is that >> the OVS flow key is not quite long enough. > > > We did not try with crc32q yet because we had the very same concerns. > In the general case, when the length of the key is not a multiple of 8 > bytes, we will have to use the crc32q to process multiple 64 bit numbers, > then process the the rest with a crc32l. My guess is that the > performance benefit won't be that high.
On 64-bit platforms, OVS already enforces alignment and multiples of 8 bytes for flow keys. If we introduced a hash primitive that does 8 bytes at a time instead of 4, there should be nothing extra to process (although I don't know how big the upside would be). _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
